-Enveda, a biotechnology company using AI to translate nature’s chemistry into new medicines, announced results from its collaboration with Anthropic to streamline critical documentation processes in drug development. Working with Claude, Enveda developed AI-enabled workflows to review Good Manufacturing Practice records and draft significant portions of an Investigational New Drug application while maintaining expert oversight and existing quality controls.

The collaboration focuses on an essential part of bringing a medicine to patients: preparing and reviewing regulatory and quality documentation. Drug-development teams review thousands of pages of manufacturing, laboratory, and clinical records, with every figure accurately documented and traceable to its source. These processes can take months and constrain even highly productive drug-discovery platforms.

“Enveda was founded to search nature at a scale that was previously impossible and turn its chemistry into better medicines faster,” said Viswa Colluru, Founder and CEO at Enveda. “Our work with Anthropic shows that AI can also help remove downstream bottlenecks in development. By giving scientists tools that make complex regulatory work faster and more reliable, we can spend more time advancing medicines for patients.”

Working with AI to Support the Quality Review of Drug Batch Records

Enveda developed the CMC Document Processor, or CDP, an application built around Claude that flags possible issues for the company’s reviewers, who decide what to do about each one. The CDP understands document types, their relationships, applicable technical and regulatory requirements, and the history of a development program. It uses this context to provide inline reviews designed to reflect what an auditor would look for to identify issues that could place a program at risk.

Enveda evaluated the CDP using 1,061 pages of batch records from a Phase 2 manufacturing campaign. Four internal reviewers and a team of experienced external consultants independently reviewed the same records, after which a member of Enveda’s Quality team adjudicated every comment based on whether an auditor would be expected to raise the issue.

Across its runs, the CDP identified all 11 actionable issues found during the human review. One timed run also raised seven additional issues relevant to an auditor, which Enveda's Quality department resolved with the manufacturer before the batch was released. These included a process step that exceeded its agreed limit and a test marked “FAIL” without a recorded result.

The system made one error after misreading a handwritten note, but appropriately classified the interpretation as requiring verification rather than presenting it as fact. Across the evaluation, Enveda’s review did not detect any hallucinated findings in the CDP.

The timed run also reduced total review time from almost 26 hours for Enveda’s four reviewers to just under three hours, including human adjudication, representing an 89% reduction for this use case. Enveda's Quality team still carries out its own final review of every record before any batch is released.

In a separate program, the CDP identified the rapid accumulation of an impurity in a non-GMP demonstration drug batch, an issue human reviewers had missed. Detecting such risks earlier can help prevent batch loss and avoid clinical delays, which can be costly.

A preprint documenting Enveda’s CDP process can be found here: https://enveda.com/cmc-document-processor/

Building Regulatory Applications from a Single Source of Truth to Accelerate IND Generation

Enveda also worked with Claude to prepare significant portions of Module 2 of an Investigational New Drug application. Rather than generating the application directly from study reports, the workflow first extracted structured facts from toxicology, bioanalytical, and pharmacokinetic source documents into a shared data layer.

This single source of truth let Enveda use consistent, traceable information throughout the application. When a source fact or analysis required refinement, the team could update it across every relevant section of the draft.

To check the work, Enveda compared Claude's draft with a Module 2 its team had already written by hand. After reviewing the draft, Enveda's Drug Development and Quality teams deemed it comparable to work produced by professional medical writers, and would have felt comfortable submitting it had the human-written version not already been sent to the publishers. Claude generated the first draft in four hours, compared with approximately three months to write a comparable first draft manually at Enveda. Neither figure includes time for expert review.

In parallel, Claude drafted an Investigator’s Brochure using the same underlying source materials. Enveda reviewers only identified one minor unit-conversion error, and after fixing that error, Enveda submitted the Investigator’s Brochure to the FDA as part of the full IND.

"We believe AI has the potential to accelerate drug development. To do this though, bottlenecks in the process, from manufacturing to clinical trials, must be addressed. Enveda has demonstrated how Claude can help people address the critical but often time consuming part of documenting each part of the work, and preparing it for final review and approval. By building these AI approaches to work, people can focus on the research that can help treat patients faster," said Aris Theologies, Lead of Life Sciences at Anthropic.

In both projects, Claude supported Enveda’s experts rather than replacing established quality processes. The system never wrote directly to a regulated GxP record. Enveda personnel reviewed and approved every finding and draft, while the company’s validated document-management system remained the system of record.